CN108431809A

CN108431809A - Use the cross-language search of semantic meaning vector

Info

Publication number: CN108431809A
Application number: CN201680074841.4A
Authority: CN
Inventors: 塞尔秋克·科普鲁; 刘明宽; 叶夫根厄·马图索夫; 哈桑·沙瓦夫
Original assignee: eBay Inc
Current assignee: eBay Inc
Priority date: 2015-12-21
Filing date: 2016-12-20
Publication date: 2018-08-21
Also published as: US20170177712A1; KR20180095910A; WO2017112650A1

Abstract

It discloses based on the member record recorded come the system and method for cluster process.Server system receives the search inquiry of first language.Server system generates semantic meaning vector associated with search inquiry.Server system accesses multiple semantic meaning vectors associated with item record, and wherein at least some of item record is not write with first language.For vectorial with the associated each corresponding semantic meaning of item record, server system compares semantic meaning vector sum with the associated semantic meaning vector of search inquiry, and is based on comparing selection item record.For selected each item record, server system determines whether the item record is write with first language, and if it is, item record is translated as first language automatically.One or more selected item records are sent to FTP client FTP for display by server system.

Description

Use the cross-language search of semantic meaning vector

Cross reference to related applications

This application claims the U.S. Patent application No.15/179 submitted on June 10th, 2016,314 priority is wanted Ask the U.S. Provisional Patent Application No.62/270,489 submitted on December 21st, 2015, on 2 11st, 2016 U.S. submitted On 2 11st, the temporary patent application No.62/293,922 and 2016 U.S. Provisional Patent Application No.62/294,060 submitted Priority.These applications are incorporated herein by reference in their entirety.

Technical field

Embodiment of the disclosure relates generally to across language on-line search, and more specifically, but is related to without limitation It is modified to transboundary search for the real-time machine translation of purposes.

Background technology

The rise of electronics and digital equipment and technology has rapidly changed society and media interact and consumer lines kimonos The mode of business.The digital technology He Zhoujie that enables people to cross the boundary of a country quickly and effectively is connected each other.However, although being easy to carry out Contact, but language difference often interferes with user and effectively associates.One such field is search and commercial field.

One solution of aphasis is the automatic machine translation for communication, search, product list etc..But Such translation needs vast resources, and translation result is often poor.

Description of the drawings

Each appended attached drawing illustrates only disclosure example embodiment, and cannot be construed to limit its range.

Fig. 1 is each functional unit for including network-based business system shown according to some example embodiments The network of client-server system environment.

Fig. 2 is the block diagram for further illustrating the FTP client FTP according to some example embodiments.

Fig. 3 is the block diagram for further illustrating the network-based business system according to some example embodiments.

Fig. 4 is the block diagram for showing the multilingual search system according to some example embodiments.

Fig. 5 is shown according to some example embodiments for executing one-stage search algorithm using semantic meaning vector and turning over The flow chart for the method translated.

Fig. 6 A- Fig. 6 C are shown according to some example embodiments for being searched using semantic meaning vector to execute single step The flow chart of rope and the method for translation.

Fig. 7 is the exemplary block diagram for showing the software architecture that may be mounted on machine according to some example embodiments.

Fig. 8 shows that the diagram of the machine of the form with computer system according to example embodiment indicates, described In computer system, one group of instruction can be executed so that any one or more in the method that machine execution is discussed herein Method.

Specific implementation mode

System, method, technology, the sequence of instructions of the illustrated examples embodiment including embodying published subject is described below Row and computing machine program product.In the following description, for purposes of explanation, many details are elaborated to provide to this The understanding of the various example embodiments of open theme.However, it will be obvious to one skilled in the art that present subject matter Embodiment may be practiced without these specific details.Usually, it is not necessary to it is real to be shown specifically well-known instruction Example, agreement, structure and technology.

Network-based business system allows user to buy commodity and service by computer network.These commodity and service Usually for people's use using many different languages in many countries.Sale multiple product and service it is network-based In business system, user can be using search inquiry Ergodic Theory to find the thing that they are look for.

However, if the language of user is different from language used in product or Service Description, translation search is needed to look into Inquiry and explanation.For example, search inquiry can be translated into desired language by system first, executed using the search inquiry after translation The source language of user is then translated into the item record (for example, product list) by the search to the item record of the language For display.This system is related to multiple translation steps, and each step can bring additional complexity to system.

On the contrary, network-based business system receives the search inquiry of first language.Network-based business system will search Rope inquiry is converted to semantic meaning vector, rather than by query translation at another language.

Each semantic meaning vector is made of the multiple values for representing one or more attributes of search inquiry.Conversion is to pass through For established model come what is completed, which has used artificial intelligence technology (for example, neural network etc.) and past user Data are trained, and semantic meaning vector field homoemorphism type is accurately created from search inquiry to create.

In some example embodiments, when product is listed in network-based business system, network-based commercial affairs Each item record is converted into semantic meaning vector by system using training pattern, and semantic meaning vector is stored in based on net In semantic meaning vector data library at the business system of network, the source language without considering item record.

Therefore, when receiving search inquiry, by search inquiry be converted into semantic meaning vector and by its be associated with The semantic meaning vector data library of item record is compared.Then, network-based business system is based on and item record phase Matching degree between the semantic meaning vector of associated each semantic meaning vector sum search inquiry is come pair and item record phase Associated each semantic meaning vector is given a mark or is sorted.In some example embodiments, it can calculate apart from score.

In some example embodiments, semantic meaning vector data library associated with item record be organized into one or The grouping of multiple themes, and network-based business system only select one theme to be grouped be compared with search inquiry (with Prevent too many unnecessary calculating).Semantic meaning vector associated with search inquiry is only and language associated with item record The finite aggregate of adopted meaning vector is compared.

When one or more item records are identified as the best match of search inquiry (based on more each item record Semantic meaning vector), network-based business system determines whether the item record uses language identical with search inquiry.It is right In being confirmed as any item record with the language different from search inquiry, network-based business system remembers the project Language appropriate is translated into record.

Then all items record is sent to client (for example, calculating associated with the user of search inquiry is submitted Machine system) for display.In some example embodiments, user selects and buys one in returning the result.In some examples In embodiment, then further improve the semantic meaning vector for creating search inquiry and item record using the purchase events Model.

Fig. 1 is each functional group for including network-based business system 120 shown according to some example embodiments The network of the client-server system environment 100 of part.Client-server system environment 100 includes at least client system System 102 and network-based business system 120.One or more communication networks 110 interconnect these components.Communication network 100 Can be any one of various network types, including LAN (LAN), wide area network (WAN), wireless network, wired network, interconnection The combination of net, personal area network (PAN) or these networks.

In some example embodiments, FTP client FTP 102 is electronic equipment, such as personal computer (PC), on knee Computer, smart phone, tablet computer, mobile phone or any other electronic equipment that can be communicated with communication network 110.Client End system 102 includes the one or more client applications 104 executed by FTP client FTP 102.In some example embodiments, Client application 104 include by search for application, communications applications, productivity application, game application, text processing application or it is any its The application of one or more of his the useful group of application composition.Client application 104 includes web browser.FTP client FTP 102 send request to network-based business system 120 using web browser and are connect from network-based business system 120 Request is received, and shows the information received from network-based business system 120.

In some example embodiments, FTP client FTP 102 includes that custom-made is used for and network-based business system 120 applications (for example, iPhone applications) communicated.In some example embodiments, network-based business system 120 is System associated with one or more services.

In some example embodiments, FTP client FTP 102 sends to network-based business system 120 and asks, request Webpage associated with network-based business system 120.For example, user is logged on to using FTP client FTP 102 based on network Business system 120, and to network-based business system 120 submit search inquiry.In response, network-based commercial affairs system System 120 generates the list one or more item records of search inquiry (for example, matching) of search results, and to FTP client FTP 102 return to item record.FTP client FTP 102 receives item record data (for example, data of the one or more products of description), And by the data include in the user interface of FTP client FTP 102.

In some example embodiments, as shown in Figure 1, network-based business system 120 is typically based on three-decker, by Front end layer is formed using logical layer and data Layer.Technical staff in correlation computer and internet arena is appreciated that Fig. 1 Shown in each module or engine indicate one group of executable software instruction and for executing instruction corresponding hardware (for example, Memory and processor).In order to avoid unnecessary details, it is omitted in Fig. 1 for understanding various example embodiments simultaneously Not closely related modules and engine.However, it will be readily appreciated by those skilled in the art that each additional function modules It can be used together with network-based business system 120 with engine, it is all as shown in Figure 1, to promote not having herein There is the additional function being particularly described.In addition, in various arrangements, each function module and engine can be resident shown in Fig. 1 On individual server computer, or it can be distributed between several server computers.Moreover, although being retouched in Fig. 1 It is depicted as three-decker, but it is not limited to this structure for various example embodiments.

As shown in Figure 1, front end is made of interface module (for example, network server) 122, the interface module 122 is from each FTP client FTP 102, which receives, to be searched for and sends search result to FTP client FTP 102 appropriate.In some example embodiments In, interface module 122 realizes that single application programming interfaces (API), all clients system 102 are connect using the single application program It mouthful sends search inquiry and receives search result.

As shown in Figure 1, data Layer includes several databases, include the use for storing network-based business system 120 The database of the various data at family, various data include historical trading data 130 and list vector data 134.

In some example embodiments, historical trading data 130 includes the data of description the following terms：Past user buys Data, it is input by user for initiates buy search inquiry, be shown to user search result and with particular transaction phase Any other data closed.In some example embodiments, 120 usage history transaction data 130 of network-based business system To develop for creating semantic meaning vector field homoemorphism type from item record and search inquiry.In some example embodiments, history Transaction data 130 further includes each search inquiry and the language of item record so that network-based business system 120 will search for Inquiry is associated with the item record of different language.

In some example embodiments, list vector data includes semantic meaning vector data library, wherein each semanteme contains Adopted vector is associated with specific project record.In some example embodiments, semantic meaning vector includes by computer learning mould A series of values (for example, it may be possible to being hundreds of values) that type generates.In some example embodiments, database is based on using for responding It the data (for example, for searching index of search result) of family search inquiry and indicates past search inquiry and is showing The data of any user interaction (for example, user clicks) generated after search result carry out tissue.Therefore, network-based commercial affairs System 120 can identify what which search entry caused to record specific project using the data about search result of storage It clicks and buys.

In some example embodiments, list vector data 134 is organized into classification, group, product classification etc..With this side Formula, network-based business system 120 can restrict the search to specific product category to improve efficiency.

Network-based business system 120 can provide extensive other application and service so that user has an opportunity to buy With the information etc. for selling article, shared and reception is customized generally according to the interest of user.

In some example embodiments, include various application server modules using logical layer, these module combination interfaces Module 122 receives the user search queries from various FTP client FTPs (102) and search result is returned to these FTP client FTP 102.

In some example embodiments, can also include vector generation module 124 and Vectors matching in application logical layer Module 126.It certainly, can be respectively at it using the other application or service of vector generation module 124 and Vectors matching module 126 Realize in the application server module of oneself.

As shown in Figure 1, in some example embodiments, vector generation module 124 and Vectors matching module 126 are implemented as The service operated in conjunction with various application server modules.For example, any amount of server module that is used alone can Call the function of vector generation module 124 and Vectors matching module 126.However, in various alternative example embodiments, vector is raw The application server module of themselves is may be implemented as at module 124 and Vectors matching module 126 so that they are as only Vertical application is run.

In general, the reception of vector generation module 124 includes the searching request of search inquiry.In some example embodiments, to The search inquiry received is converted to semantic meaning vector by amount generation module 124.In some example embodiments, semantic meaning Vector is generated based on the model that usage history transaction data 130 is trained, with the logical of each item record of determination and search inquiry Use attribute.In some example embodiments, with the generation of New Transaction, the more new model of vector generation module 124 is to be incorporated to new number According to.In some example embodiments, the item record of different language and search inquiry can be converted into general semantics and contain by model Adopted vector so that can be compared them in the case where not considering their language.

Similarly, when receiving new item record from FTP client FTP 102, vector generation module 124 creates this The semantic meaning vector of program recording.Then the semantic meaning vector newly created is stored in list records vector data 134. In some example embodiments, vector generation module 124 determines product category associated with item record and is based on product class Do not carry out the association semantic meaning vector in organized list vector data 134.

In some example embodiments, list vector data 134 has established product category hierarchical structure, and every A item record is placed in one or more classifications of hierarchical structure.

The semantic meaning vector that Vectors matching module 126 is created using vector generation module 124 for particular search query To search the matching of the search inquiry in list vector data 134.In some example embodiments, Vectors matching module 126 will The semantic meaning vector of search inquiry is compared with each semantic meaning vector being stored in list vector data, and raw At the matching score of each semantic meaning vector.

In some example embodiments, Vectors matching module 126 generates the distance between two semantic meaning vectors score (similarity between two semantic vectors is wherein indicated apart from score).Then, Vectors matching module 126 is based on associated obtain Divide and each item record semantic meaning vector is ranked up.

In some example embodiments, Vectors matching module 126 determines the specific quantity of required item record result, and And based on the item record semantic meaning vector of the sequencing selection quantity.For selected each semantic meaning vector, vector Matching module 126 receives associated item record, and if desired, item record is translated into the use for submitting search inquiry The language at family is (for example, translate into the language or translate by the another of user's instruction of submission search inquiry that search inquiry uses Kind language).

In some example embodiments, then, selected item record is sent to FTP client FTP 102 for aobvious Show.

Fig. 2 is the block diagram for further illustrating the FTP client FTP 102 according to some example embodiments.Client device 102 generally include one or more central processing unit (CPU) 202, one or more network interfaces 210, memory 212 and use In the one or more communication bus 214 for interconnecting these components.Client device 102 includes user interface 204.User interface 204 include display equipment 206, and optionally includes the input of such as keyboard, mouse, touch-sensitive display or other input units Equipment 208.In addition, some client devices 102 carry out supplement or alternative other input equipments using microphone and speech recognition.

Memory 212 includes such as dynamic random access memory (DRAM), static RAM (SRAM), double The high random access storage of data rate random access memory (DDR RAM) or other arbitrary access fixed memory devices etc. Device；And may include such as one or more disk storage equipments, optical disc memory apparatus, flash memory device or other are non-volatile The nonvolatile memory of solid storage device etc..Memory 212 can optionally include the one or more far from CPU 202 Storage device.Memory 212 (or alternatively, the non-volatile memory device in memory 212) includes that non-transitory calculates Machine readable storage medium storing program for executing.

In some example embodiments, the following journey of the computer-readable recording medium storage of memory 212 or memory 212 Sequence, module and data structure or its subset：

● operating system 216 includes the mistake for handling various basic system services and for executing hardware dependent tasks Journey；

● network communication module 218, for via one or more network interfaces 210 (wired or wireless) and one or more FTP client FTP 102 is coupled to other calculating by a communication network 110 (for example, internet, other WAN, LAN, MAN etc.) Machine；

● display module 220, the information for enabling operating system 216 and client application 104 to generate are in visually Now in display equipment 206；

● one or more client application modules 222, for handling with network-based business system (for example, in Fig. 1 System 120) interaction various aspects, including but not limited to：

ο browser applications 224 are used for network service request information associated with network-based business system 120 (for example, content item and item record) and receive sound from network service associated with network-based business system 120 It answers；And

Client data module 230, for storing the data about client, including but not limited to：

ο client profile datas 232, for storing about network-based commercial affairs associated with FTP client FTP 102 The profile data of the user of system 120.

Fig. 3 is the block diagram for further illustrating the network-based business system 120 according to some example embodiments.It is based on The business system 120 of network generally includes one or more CPU 302, one or more network interfaces 310,306 and of memory One or more communication bus 308 for interconnecting these components.Memory 306 include such as DRAM, SRAM, DDR RAM or The high-speed random access memory of other random access solid state storage devices etc.；And may include such as one or more disks The nonvolatile memory of storage device, optical disc memory apparatus, flash memory device or other non-volatile solid-state memory devices etc..It deposits Reservoir 306 can optionally include one or more storage devices far from CPU 302.

Memory 306 (or alternatively, the non-volatile memory device in memory 306) includes non-transitory computer Readable storage medium storing program for executing.In some example embodiments, the computer-readable recording medium storage of memory 306 or memory 306 Following procedure, module and data structure or its subset：

● operating system 314 includes the mistake for handling various basic system services and for executing hardware dependent tasks Journey；

● network communication module 316, for via one or more communications network interfaces 310 (wired or wireless) and one Or multiple communication networks 110 (for example, internet, other WAN, LAN, MAN etc.) couple network-based business system 120 To other computers；

● one or more server application modules 318 are configured as executing the offer of network-based business system 120 Service, including but not limited to：

ο vector generation modules 124 are based on history for search inquiry and item record to be converted to semantic meaning vector 130 training vector of transaction data generates model, and from FTP client FTP (for example, FTP client FTP 102 in Fig. 1) receiving item Program recording and search inquiry；

ο Vectors matchings module 126, for semantic meaning vector sum associated with the search inquiry received to be stored in Multiple semantic meaning vectors associated with item record in list vector data 134 are compared, and based on this comparison Select the item record of best match；

ο receiving modules 322, for being received from user via FTP client FTP (for example, FTP client FTP 102 in Fig. 1) Search inquiry and item record；

ο list blocks 324, information for being submitted based on user create item record so as to via network-based quotient Business system 120 sells product；

ο translation modules 326, for item record to be translated as second language from first language automatically；

ο language determining module 328, for record of identifying project language whether the language phase with the search inquiry of submission Together；

ο sorting modules 330, for each semantic meaning vector sum search inquiry based on item record semantic meaning to The matching degree of amount is ranked up each semantic meaning vector of item record；

ο selecting modules 332, for selecting one based on the sequence of semantic meaning vector associated with each item record A or multiple item records；

ο transmission modules 334, for selected item record to be sent to FTP client FTP (for example, the client in Fig. 1 End system 102) for display；And

ο spacing modules 336, for based on the calculating for determining the distance between two vectors determine two semantic meanings to Similarity between amount；And

● server data module 340, storage and 120 relevant data of network-based business system, including but it is unlimited In：

ο historical trading datas 130, including data of the past interaction of description (for example, sale and/bid) and about this Interactive information a bit includes for initiating the search inquiry of transaction, the search result of display and user before completing to merchandise The item record of click；And

ο lists vector data 134, the semantic meaning vector for storing multiple item records, which will It is used when being matched with the search inquiry of another language.

Fig. 4 is the block diagram for further illustrating the multilingual search system 400 according to some example embodiments.According to some Example embodiment, user connect via network with multilingual search system 400.User submits the search inquiry 410 of first language.

In some example embodiments, multilingual search system 400 is network-based business system (for example, in Fig. 1 System 120) component and receive the search inquiry from user 402.In some example embodiments, vector generation module 124 Receive search inquiry.Vector generation module 124 creates the semantic meaning vector 412 of search inquiry, rather than search inquiry is turned over It is translated into other one or more language.

In some example embodiments, vector generation module 124 includes mapping queries to semantic meaning vector field homoemorphism type. In some example embodiments, usage history transaction data 130 carrys out training pattern.In some example embodiments, using such as The computer learning technology of decision tree learning, artificial neural network and depth learning technology, support vector machines, Bayesian network etc. To build model itself.

For example, vector generation module 124 identifies the item record (for example, product) of the search and second language of first language Purchase between all historical tradings.

In some example embodiments, the semantic meaning vector 412 of search inquiry 410 is created in vector generation module 124 Later, which is transferred to Vectors matching module 126.

In some example embodiments, Vectors matching module 126 analysis be stored in list vector data 134 and with one Or multiple associated multiple semantic meaning vectors of item record, the semantic meaning created from search inquiry 410 with identification matching One or more semantic meanings vector of vector 412.In some example embodiments, each semantic meaning vector 412 includes more A value, and Vectors matching module 126 creates and indicates that each semantic meaning vector sum associated with item record is looked into search Ask the score of the similarity between 410 associated semantic meaning vectors 412.

After all semantic meaning vectors 412 in having evaluated list vector data 134,126 base of Vectors matching module One or more semantic meaning vectors 412 are selected from list vector data 134 in the score of generation.Then, with it is selected Semantic meaning 412 associated item records of vector are sent to machine translation module 408.In some example embodiments, machine Device translation module 408 determines that the language of which of selected item record (if any) is different from and search inquiry 410 associated language.

If the language of any of selected item record is confirmed as the language different from search inquiry 410, Item record is translated as searching by machine translation module 408 from their source language (for example, language that they are submitted) automatically The language of rope inquiry 410.Then, the item record 414 after translation is sent to FTP client FTP (for example, the client in Fig. 1 System 102).

Fig. 5 is shown according to some example embodiments for executing one-stage search algorithm using semantic meaning vector and turning over The flow chart for the method 500 translated.Each operation shown in fig. 5 can correspond to be stored in computer storage or computer-readable Instruction in storage medium.In some embodiments, the method 500 described in Fig. 5 by network-based business system (for example, System 120 in Fig. 1) it executes.However, method 500 can also be executed by any other suitable configuration of electronic hardware.

In some embodiments, method 500 is at including network-based business system (for example, system 120 in Fig. 1) It executes, network-based business system includes one or more processors and memory, and memory is stored for by one or more One or more programs that a processor executes.

In some example embodiments, network-based business system (for example, system 120 in Fig. 1) is from client system System (for example, FTP client FTP 102 in Fig. 1) receives (502) search inquiry (for example, search inquiry 410).Search inquiry has Associated first language (for example, writing the language of search inquiry).

In some example embodiments, network-based business system (for example, system 120 in Fig. 1) is based on past Data are bought, model is created using computer learning model.In some example embodiments, using deep learning or nerve net Network learning method creates model.In some example embodiments, network-based business system is (for example, the system in Fig. 1 120) (504) semantic meaning vector (for example, semantic meaning vector 412) associated with search inquiry is created using model. For example, network-based business system (for example, system 120 in Fig. 1) creates a model, which uses search inquiry Text carry out generative semantics meanings vector.Semantic meaning vector is to indicate the search inquiry in multi-C vector space or item record Position (for example, wherein position is based on semantic meaning) a series of numbers.

In a very simple example, for two-dimensional space of (x, the y) value between 0 to 1, model is trained To indicate the different zones with different semantic meanings in two-dimensional space.Then, each item record and search inquiry can be by It is right that model is mapped to specific (x, y).Then, network-based business system (for example, system 120 in Fig. 1) passes through calculating The similarity between search inquiry and item record is determined in the space (x, y) at the distance between 2 points.

Usually, semantic meaning vector will be mapped to that in the vectors of hundreds of dimensions so that model can indicate extremely complex Semantic meaning.

In some example embodiments, model using past search inquiry whole corpus and with they caused by Purchase associated item record identify the semantic relation between inquiry and item record.In some example embodiments, Relationship can be based on entry the common frequency of occurrences (for example, under the sufficiently large document scale of construction, determine which entry in phase identical text Occur in shelves, model can be made to efficiently generate semantic meaning vector).In some example embodiments, the importance of entry is logical It crosses and is weighted against word frequency score.

In other example embodiments, determine that semantic dependency carrys out training pattern by using neural network.In the example In, neural network receive input (for example, about search inquiry or the various data of item record, including text, sending time, Position sources etc.).Each in these inputs is endowed a weight and passes to multiple concealed nodes.Concealed nodes exchange letter It ceases (weight of imparting), to generate output.In some example embodiments, there are several concealed nodes layers.In this case Output be multi-C vector.For example, first semantic meaning vector may include smv₁=(v₁, v₂, v₃, v₄..., v_n) in The list of value.

In some example embodiments, it is trained using available data (for example, with matched search inquiry is successfully bought) Model, and Learning Algorithm is adaptively adjusted weight to generate the inquiry and item record of matching existing record Semantic meaning vector.In some example embodiments, when New Transaction occurs, carry out more new model using new data.

In some example embodiments, semantic meaning vector also creates the semantic meaning of search inquiry using its dependent variable Vector, dependent variable include the feature and history for submitting user, the time of search inquiry and position etc..

After generating the semantic meaning vector of search inquiry, network-based business system (for example, be in Fig. 1 System is 120) by the semantic meaning vector sum of search inquiry semantic meaning vector associated with multiple item records (for example, each Item record, which all has, is stored in being associated in the database of network-based business system (for example, system 120 in Fig. 1) Semantic meaning vector) compare (506).

In some example embodiments, network-based business system (for example, system 120 in Fig. 1) calculates two languages Similarity score between adopted meaning vector or degree of approach score.In some example embodiments, similarity score is n dimension Euclideans (Euclidean) distance.In other example embodiments, Chebyshev (Chebyshev) distance, Hamming can be used (Hamming) distance, geneva (Mahalanobis) distance, Manhattan (Manhattan) distance, Minkowski (Minkowski) distance, Ha Weixin (Haversine) distances or any other appropriately distance calculate to calculate score.

In some example embodiments, network-based business system (for example, system 120 in Fig. 1) is from being stored in base (508) one or more projects are selected in multiple item records of the business system (for example, system 120 in Fig. 1) of network Record.The similarity score or the degree of approach of semantic meaning vector associated with item record determined by being based at least partially on Score selects item record.As set forth above, it is possible to calculate the distance between two vectors using various algorithms.Calculating has A specific example of the similarity between two vectors of t vector value is following (still, as set forth above, it is possible to use in total Many different algorithms)：

Therefore, it calculates with the query semantics meaning vector sum item record semantic meaning vector of t vector value in total Between similarity.The calculating will produce a value.Inquiry is lower with this value of item record pair, then in vector space, looks into Inquiry is confirmed as closer with item record.

In some example embodiments, item record database includes the item record from a variety of different languages.It is semantic Meaning vector makes the meaning between language standardize.Therefore, once it is determined that one or more item records, network-based quotient Item record translation (510) is if necessary first language by business system (for example, system 120 in Fig. 1).In this way, First user can use their language to submit search inquiry, and obtain the result of the product of other language.

Once item record is translated into language appropriate, network-based business system is (for example, the system in Fig. 1 120) item record after translation is sent into (512) to FTP client FTP (for example, FTP client FTP 102 in Fig. 1) for aobvious Show.

Fig. 6 A be show according to some example embodiments for using semantic meaning vector come execute one-stage search algorithm and The flow chart of the method for translation.Each operation can correspond to be stored in computer storage or computer-readable shown in Fig. 6 A Instruction in storage medium.Can selection operation indicated by the dotted line frame of dotted border (for example, with).In some embodiments, Method described in Fig. 6 A is executed by network-based business system (for example, system 120 in Fig. 1).However, the method It can be executed by any other suitable configuration of electronic hardware.

In some embodiments, method 600 is at including network-based business system (for example, system 120 in Fig. 1) It executes, network-based business system includes one or more processors and memory, and memory is stored for by one or more One or more programs that a processor executes.

In some example embodiments, network-based business system (for example, system 120 in Fig. 1) receives (602) item Program recording, to be contained in network-based business system (for example, system 120 in Fig. 1).Item record is will be The description for the product sold on network-based business system (for example, system 120 in Fig. 1).Description may include title, production Product specification and feature, image and any other relevant information.

Large-scale network-based business system (for example, system 120 in Fig. 1) can make its service can be by many countries User use so that user says multilingual perhaps.Therefore, item record may be and almost be write with any language.In order to Item record is set to standardize, network-based business system (for example, system 120 in Fig. 1) (604) generates the project received The semantic meaning vector of record.As described above, semantic meaning vector is a series of digital (or the values) for indicating item record feature. In some example embodiments, network-based business system (for example, system 120 in Fig. 1) uses past transaction data And computer learning technology (for example, neural network) is combined to create the semantic meaning that will generate item record and search inquiry Vector field homoemorphism type.In some example embodiments, all there are one models for each possible language.Therefore, any item record or Search inquiry will all be converted into the semantic meaning vector that can be compared, and without considering original language.

As described above, by using the model of (above-mentioned example uses neural network) training in any suitable manner, project Record and search inquiry are converted into semantic meaning vector, which uses data conduct associated with inquiry or item record Input and generate the semantic meaning vector of n dimensions (often hundreds of dimensions).By using a large amount of completed transaction data, wherein at least Some transaction include carried out and eventually led to first language for the item record created with second language project it is complete It, can be associated with the item record of second language by the inquiry of first language with training pattern at the search of transaction.Once generating Such model (for example, using training data come adjust neural network input weight and hiding weight to generate accurately Semantic meaning vector), the search inquiry of first language can be converted into and can be carried out with the semantic meaning vector of second language The semantic meaning vector compared, without translation and inquiry or item record.

In some example embodiments, network-based business system (for example, system 120 in Fig. 1) analysis (606) with The item record of semantic meaning vector correlation connection is to identify the product category joined with semantic meaning vector correlation.For example, if production Product are a pair of shoes, then product classification can be footwear by network-based business system (for example, system 120 in Fig. 1).Such as with It is lower discussed, there is the product category joined with each item record and semantic meaning vector correlation can improve search efficiency.

In some example embodiments, product category is determined based on the analysis to search inquiry.For example, network-based Business system (for example, system 120 in Fig. 1) includes the database of entry and matched product category.Implement in other examples In example, user selects specific products classification when submitting search inquiry.In other example embodiments, first by search inquiry Semantic meaning vector is compared with a series of semantic meaning vectors of multiple product categories are indicated.By immediate matching (example Such as, using above-mentioned algorithm) it is determined as product category associated with the inquiry.

In some example embodiments, network-based business system (for example, system 120 in Fig. 1) tissue (608) number According to library so that each semantic meaning vector is associated with identified product category.

In some example embodiments, network-based business system (for example, system 120 in Fig. 1) is by semantic meaning Vector storage (610) is in the database of network-based business system (for example, system 120 in Fig. 1).Therefore, Mei Gexiang Program recording is indicated by the semantic meaning vector in database.In some example embodiments, by product tissue database.

It should be noted that operation 602-610 describes the database for creating content item vector and establishment is trained to Generate the process of the semantic meaning vector field homoemorphism type of search inquiry and contents of a project record.These steps can need mould receiving It is whenever performed off before the inquiry of type and database.Therefore, although Fig. 6 A show operation 612 immediately in operation After 610, but there may be the plenty of time between the two operations.

Operation 612 is in response in receiving search inquiry and a part for generative semantics meaning vector in real time.Therefore, it operates The step of being indicated in 602-610 completes some time point before the generation of real-time semantic meaning vector, but need not thing to working as It is preceding just to complete.

In some example embodiments, network-based business system (for example, system 120 in Fig. 1) is from client system System (for example, FTP client FTP 102 in Fig. 1) receives the search inquiry of (612) first language.In some example embodiments, User submits search inquiry with the text of its preferred language.In some example embodiments, network-based commercial affairs system System (for example, system 120 in Fig. 1) automatically detects search inquiry based on text, the source place of search inquiry and user characteristics Language.

In some example embodiments, network-based business system (for example, system 120 in Fig. 1) generate (614) with The associated semantic meaning vector of search inquiry.As described above, network-based business system (for example, system 120 in Fig. 1) The model trained using the machine learning algorithm using such as neural network.The training, which uses, comes from network-based business system The historical data (for example, purchase and click and the search inquiry for initiating these interactions) of (for example, system 120 in Fig. 1). In some example embodiments, each language use language particular model carrys out generative semantics meaning vector.

In other example embodiments, for each original language/object language to using different models.Therefore, such as Fruit supports three kinds of language (language A, language B and language C), then may have six kinds of models (for example, a kind of model looking into language A Ask the item record for being matched to language B, a kind of model incites somebody to action the item record of the match query of language A to language C, a kind of model The match query of language B to language A item record, it is such).

In some example embodiments, network-based business system (for example, system 120 in Fig. 1) identification (616) with The associated first language of search inquiry.

Fig. 6 B are further illustrated according to some example embodiments for executing single step using semantic meaning vector The flow chart of the method 600 of search and translation.Shown in Fig. 6 B each operation can correspond to be stored in computer storage or Instruction in computer readable storage medium.Can selection operation indicated by the dotted line frame of dotted border (for example, with).At some In embodiment, the method described in Fig. 6 B is executed by network-based business system (for example, system 120 in Fig. 1).However, The method can also be executed by any other suitable configuration of electronic hardware.

In some example embodiments, network-based business system (for example, system 120 in Fig. 1) selection (618) with The associated semantic meaning vector of first language identified generates model.Then, network-based business system is (for example, Fig. 1 In system 120) using selected semantic meaning vector generate model come generate search inquiry semantic meaning vector.

In some example embodiments, network-based business system (for example, system 120 in Fig. 1) access (622) with Multiple associated multiple semantic meaning vectors of item record, wherein at least some of item record is compiled with first language It writes.For example, network-based business system (for example, system 120 in Fig. 1) storage creates when submitting item record with user Each item of articles associated semantic meaning vector data library.

In some example embodiments, item record is write with a variety of different languages.In some example embodiments, project Record may include the content using any language.

In some example embodiments, accessing multiple semantic meaning vectors associated with item record includes：Based on net The business system (for example, system 120 in Fig. 1) of network analyzes (624) search inquiry to identify associated with search inquiry one A or multiple product categories.For example, search inquiry can be analyzed based on the text of search inquiry, to reduce search range.

For example, network-based business system (for example, system 120 in Fig. 1) access (626) and identify one or Multiple associated semantic meaning vectors of product category.In this way, network-based business system (for example, be in Fig. 1 120) system can limit the quantity of semantic meaning vector needed to be considered when executing search.

In some example embodiments, it for vectorial with the associated each corresponding semantic meaning of item record, is based on The business system (for example, system 120 in Fig. 1) of network is by corresponding semantic meaning vector sum language associated with search inquiry Adopted meaning vector compares (628).Compare is to determine which item record is most matched with search inquiry.

In some example embodiments, by corresponding semantic meaning vector sum semantic meaning associated with search inquiry to Amount, which compares, further includes：It is related to search inquiry that network-based business system (for example, system 120 in Fig. 1) calculates (630) Degree of approach score between the corresponding semantic meaning vector of semantic meaning vector sum of connection.

Fig. 6 C be show according to some example embodiments for using semantic meaning vector come execute one-stage search algorithm and The flow chart of the method 600 of translation.Each operation can correspond to be stored in computer storage or computer shown in Fig. 6 C Instruction in readable storage medium storing program for executing.Can selection operation indicated by the dotted line frame of dotted border (for example, with).In some embodiments In, the method described in Fig. 6 C is executed by network-based business system (for example, system 120 in Fig. 1).However, the side Method can also be executed by any other suitable configuration of electronic hardware.

In some example embodiments, network-based business system (for example, system 120 in Fig. 1) is based on associated Degree of approach score multiple semantic meaning vectors are ranked up (632).Therefore, most related (or in semantic meaning vector side Face is close) semantic meaning vector come foremost.

In some example embodiments, network-based business system (for example, system 120 in Fig. 1) is based on and project The comparison between associated semantic meaning vector sum semantic meaning vector associated with search inquiry is recorded to select (634) One or more item records.In some example embodiments, it is based at least partially on and joins with each semantic meaning vector correlation Sequence select one or more item records.In this way, maximally related item record is selected.

For selected each corresponding item record, network-based business system is (for example, the system in Fig. 1 120) determine whether (636) corresponding item record is write with first language.For example, network-based business system is (for example, figure System 120 in 1) determine search inquiry language (for example, language 1) and corresponding item record language, and comparison search The language of the language of inquiry and corresponding item record.

It is not write with first language according to the corresponding item record of determination, network-based business system is (for example, figure System 120 in 1) it is automatically first language by corresponding item record translation (638).If corresponding item record is to use What first language was write, except non-user is indicated otherwise, otherwise do not need this translation.

Then, network-based business system (for example, system 120 in Fig. 1) is by one or more selected projects Record sends (640) to FTP client FTP (for example, FTP client FTP 102 in Fig. 1) for display.

Module, component and logic

Some embodiments are described herein as including logic or multiple components, module or mechanism.Module may be constructed Software module (for example, the code embodied on machine readable media) or hardware module." hardware module " is to be able to carry out certain behaviour The tangible unit made, and can configure or arrange according to certain physics mode.It is one or more in various example embodiments Computer system (such as independent computer system, client computer system or server computer system) or computer One or more hardware modules (such as processor or processor group) of system can be matched by element (such as apply or application obscure portions) Operation is set to execute the hardware module of specific operation described herein.

In some embodiments, hardware module can mechanically, electronically or its it is arbitrary it is appropriately combined come it is real It is existing.For example, hardware module may include the special circuit or logic for being for good and all configured as executing specific operation.For example, hardware Module can be application specific processor, such as field programmable gate array (FPGA) or application-specific integrated circuit (ASIC).Hardware module is also May include by software provisional configuration be execute specific operation programmable logic or circuit.For example, hardware module may include The software executed by general processor or other programmable processors.Once by such software configuration, hardware module reforms into Specific machine (or specific components of machine), by custom-made for executing configured function, and is no longer general place Manage device.It should be understood that：Mechanically, (such as matched by software with circuit that is special and permanently configuring or with the circuit of provisional configuration Set) realize decision can be for cost and time the considerations of of hardware module.

Therefore, phrase " hardware module " is interpreted as covering tangible entity, be physically construction, permanent configuration (such as Rigid line connects) or provisional configuration (such as programming) be to operate or execute in a specific way the entity of specific operation described herein. As it is used herein, " hard-wired module " refers to hardware module.Consider the reality of provisional configuration (such as programming) hardware module Example is applied, without configuring or instantiating each in hardware module at any one time.For example, including being matched by software in hardware module In the case of being set to the general processor for application specific processor, general processor can be configured to difference in different time Application specific processor (such as including different hardware modules).Therefore, software by specific one or multiple processors for example with It is set to and constitutes specific hardware module a moment and constitute different hardware modules at another moment.

Hardware module can provide information to other hardware modules and receive information from other hardware modules.Therefore, it is retouched The hardware module stated can be counted as being communicatively coupled.It, can by two or more if existing simultaneously multiple hardware modules Signal transmission (such as passing through circuit appropriate and bus) between multiple hardware modules realizes communication.Exist in multiple hardware modules It, can be for example by storing and obtaining the addressable storage of multiple hardware modules in the embodiment that different time is configured or instantiated Information in device structure realizes the communication between such hardware module.For example, a hardware module can execute operation simultaneously The output of the operation is stored in the storage device for communicating with coupling.Another hardware module can then access memory later Equipment, to fetch and handle stored output.Hardware module can also be initiated and be inputted or the communication of output equipment, and energy It is enough that resource (such as set of information) is operated.

The various operations of exemplary method described herein can at least partly by provisional configuration (such as passing through software) or The permanent one or more processors for being configured to execute relevant operation execute.Either interim or permanent configuration, such place Reason device may be constructed operation to execute the module that the processor of one or more operations described herein or function is realized.As herein Used, " module that processor is realized " refers to the hardware module realized using one or more processors.

Similarly, method described herein can be realized by processor at least partly, wherein par-ticular processor or multiple Processor is the example of hardware.For example, what at least some operations of method can be realized by one or more processors or processor Module executes.In addition, one or more processors are also operable to support " in cloud computing environment or as " software takes Business " (SaaS) executes related operation.For example, at least some of operation can be by computer (as the machine including processor Example) group executes, and these operations can be via network (such as internet) and via one or more appropriate interfaces (such as using journey Sequence interface (API)) it accesses.

The execution of certain operations can be distributed in the processor, not only resided in individual machine, but be deployed in more In a machine.In some example embodiments, the module that processor or processor are realized can be located at single geographical location (example Such as, in home environment, working environment or server zone).In other example embodiments, the mould of processor or processor realization Block can be distributed in multiple geographical locations.

Machine and software architecture

In some embodiments, the module that describes in conjunction with Fig. 1 to Fig. 6, method, using etc. in machine and associated software frame It is realized in the context of structure.Following chapters and sections describe the representative software architecture for being applicable in and being used together with the disclosed embodiments With machine (for example, hardware) framework.

Software architecture is used together with hardware structure, to create equipment and machine for special-purpose customization.For example, with The specific hardware framework of specific software framework coupling will create mobile device, mobile phone, tablet device etc..It is slightly different Hardware and software framework can generate the smart machine for " Internet of Things ", and another combination then generates and is used for cloud computing framework Interior server computer.All combinations of such software and hardware framework are not described herein, because of art technology Personnel can easily understand that how to realize the present invention's in the different contexts different from the disclosure contained herein content Theme.

Software architecture

Fig. 7 is the block diagram 700 for showing representative software architecture 702, which can combine this paper institutes The various hardware structures of description are used together.Fig. 7 is only the non-limiting example of software architecture 702, and it should be appreciated that can be with Many other frameworks are realized to help to realize function described herein.Software architecture 702 can be in the machine of such as Fig. 8 It is executed on 800 hardware, the machine 800 further includes processor 810, memory/storage 830 and I/O components 850 etc.. Representative hardware layer 704 is shown in the figure 7, and can indicate the machine 800 of such as Fig. 8.Representative hardware layer 704 Include one or more processing units 706 with associated executable instruction 708.Executable instruction 708 indicates software architecture The realization of 702 executable instruction, including the method for Fig. 1-6, module etc..Hardware layer 704 further includes memory and/or storage mould Block 710, the memory and/or memory module 710 also have executable instruction 708.Hardware layer 704 can also include indicating hard Other hardware 712 of any other hardware (such as being illustrated as other hardware of a part for machine 800) of part layer 704.

In the exemplary architecture of Fig. 7, software architecture 702 can be conceptualized as the storehouse of layer, wherein every layer of offer is specific Function.For example, software architecture 702 may include such as operating system 714, library 716, frame/middleware 718, using 720 and Each layer of expression layer 744.Optionally, application can be called by software stack using the other assemblies in 720 and/or layer Program Interfaces (API) call 724, and response, the return value etc. for being shown as message 726 are received in response to API Calls 724 Deng.Shown layer is representative in itself, and not all software architecture all has all layers.For example, some are moved Dynamic or special purpose operating system may not provide frame/middleware layer 718, and other systems can provide such layer.It is other soft Part framework may include extra play or different layers.

Operating system 714 can manage hardware resource and provide public service.Operating system 714 may include such as kernel 728, service 730 and driving 732.Kernel 728 may be used as the level of abstraction between hardware and other software layer.For example, kernel 728 It can be responsible for memory management, processor management (for example, scheduling), assembly management, networking, security setting etc..Service 730 can be with Other public services are provided for other software layers.Driver 732 can be responsible for controlling bottom hardware or connect with bottom hardware interface It connects.For example, depend on hardware configuration, driving 732 may include display driving, camera driving,Driving, flash memory drive Dynamic, serial communication driving (such as universal serial bus (USB) driving),Driving, audio driven, power management driving Etc..

Library 716 can provide the public infrastructure that can be used by application 720 or other assemblies or layer.Library 716 usually carries For allowing other software module with than directly with the function of underlying operating system 714 (for example, kernel 728, service 730 and/or drive Dynamic 732) interface connection easier way executes task.Library 716 may include that can provide such as memory allocation function, word The system library 734 (for example, C java standard libraries) of the function of symbol string operating function, math function etc..In addition, library 716 may include API Library 736, for example, media library (for example, for support various media formats (such as MPEG4, H.264, MP3, AAC, AMR, JPG, PNG the library of presentation and manipulation)), shape library is (for example, can be used for rendering the 2D's and 3D in graphical content over the display OpenGL frames), database (for example, can provide the SQLite of various relation data library facilities), the libraries web be (for example, can carry For the WebKit of internet browsing function) etc..Library 716 can also include various other libraries 738, to provide to using 720 With many other API of other software components/modules.

Frame/middleware 718 can provide can be by the higher utilized using 720/ or other software components/modules Public infrastructure.For example, frame/middleware 718 can provide various graphic user interfaces (GUI) function, premium resource pipe Reason, high-level position service etc..Frame/middleware 718, which can provide, to be utilized by 720/ or other software components/modules of application Broad range of other API, some of them can be exclusively used in specific operation system or platform.

Include built-in application 740 or third-party application 742 using 720.The example of representative built-in application 740 can wrap Include but be not limited to contact application, browser application, book readers application, location application, media application, message transmission application Or game application.Third-party application 742 may include any built-in application 740 and various other applications.Specific In example, third-party application 742 by the entity different from the supplier of particular platform (for example, use Android^TMOr iOS^TMIt is soft Part development kit (SDK) and develop application) can be in Mobile operating system (such as iOS^TM、Android^TM、Phone or other Mobile operating systems) on the mobile software that runs.In this example, third-party application 742 It can call by the API Calls 724 of the Mobile operating system offer of such as operating system 714 etc, to help to realize herein The function of description.

Using 720 can utilize built-in operation system function (for example, kernel 728, service 730 and/or driving 732), Library (for example, system library 734, API library 736 and other libraries 738), frame/middleware 718 create user interface, with system User interact.Alternatively or additionally, in some systems, expression layer (such as table can be passed through with the interaction of user Show layer 744) occur.In such systems, the various aspects for the application/module that can be interacted with user using/module " logic " Separation.

Some software architectures utilize virtual machine.In the example in figure 7, this is shown by virtual machine 748.Virtual machine creating software Environment, wherein can be executed as being executed on hardware machine (for example, machine 800 of Fig. 8) using/module.Virtual machine by Master operating system (for example, operating system 714) trustship, and usually there is (although not always) behaviour of management virtual machine 748 The monitor of virtual machine 746 of work and the interface being connect with master operating system (for example, operating system 714).Software architecture is in void Quasi- machine 748 (for example, operating system 750, library 752, frame 754, using 756 or expression layer 758) in execute.In virtual machine 748 The layer of these software architectures of interior execution can be identical as previously described respective layer, or can be different.

Example machine framework and machine readable media

Fig. 8 be show according to some example embodiments can from machine readable media (for example, machine readable storage be situated between Matter) in read and instruct and the block diagram of the component of the machine 800 that any one or more of executes process discussed herein.Tool Body, Fig. 8 shows the schematic diagram of the machine 800 of the exemplary forms of computer system, wherein can execute instruction 816 (for example, Software, program, using, small application, app or other executable codes) so that machine 800 execute any type being discussed herein or A variety of methods.For example, instruction 816 can make the flow chart of the execution Fig. 5 of machine 800-Fig. 6.Instruction 816 will be general unprogrammed Machine 800 is transformed into the specific machine for being programmed to execute described and illustrated function in the manner described.Alternative real It applies in example, machine 800, which operates as autonomous device or can couple (for example, networking), arrives other machines.In networked deployment, machine Device 800 can in server-client network environment using the volume operation of server machine or client machine or as Peer machines operation in equity (or distributed) network environment.Machine 800 can include but is not limited to server computer, visitor Family end computer, personal computer (PC), tablet computer, laptop computer, net book, set-top box (STB), individual digital Assistant (PDA), entertainment medium system, cellular phone, smart phone, mobile device, wearable device (for example, smartwatch), Smart home device (for example, intelligent appliance), other smart machines, the network equipment, network router, the network switch, bridge, Or it can sequentially or otherwise execute any machine of the instruction 816 for the action that specified machine 800 to be taken.In addition, Although illustrating only individual machine 800, term " machine " will also be believed to comprise the set of machine 800, individually or 816 are jointly executed instruction to execute any one or more of the method being discussed herein.

Machine 800 may include the processor 810 that can be configured as example communicating with one another via bus 802, memory/deposit Store up equipment 830 and I/O components 850.In the exemplary embodiment, processor 810 is (for example, central processing unit (CPU), simplify finger Enable collection calculate (RISC) processor, complex instruction set calculation (CISC) processor, graphics processing unit (GPU), at digital signal Manage device (DSP), application-specific integrated circuit (ASIC), RF IC (RFIC), another processor or its be any appropriately combined) it can With including can for example execute instruction 816 processor 812 and processor 814.Term " processor " is intended to include that may include It may be performed simultaneously the multi-core processor of two or more independent processors (sometimes referred to as " core ") of instruction 816.Although Fig. 8 Show multiple processors 810, but machine 800 may include the single processor with single core, the list with multiple cores A processor (for example, multi-core processor), multiple processors with single core, with multiple processors of multiple cores or its What is combined.

Memory/storage 830 may include memory 832 (for example, main memory or other storage storage facilities), And both storage unit 836, memory device 932 and storage unit 936 can be visited for example via bus 802 by processor 810 It asks.Storage unit 836 and memory 832 store the instruction 816 for embodying any one or more of method or function as described herein. During machine 800 executes instruction 816, instruction 816 can also be resided in completely or partially in memory 832, storage is single In member 836, at least one of processor 810 (for example, in the cache memory of processor) or its is any suitable In combination.Therefore, the memory of memory 832, storage unit 836 and processor 810 is the example of machine readable media.

As used herein, " machine readable media " is the equipment for referring to temporarily or permanently store instruction and data, And it can include but is not limited to random access memory (RAM), read-only memory (ROM), buffer storage, flash memory, optics Medium, magnetic medium, cache memory, other types of memory are (for example, Erasable Programmable Read Only Memory EPROM Or its any suitable combination (EEPROM)).Term " machine readable media " should be considered as including the list for capableing of store instruction 816 A medium or multiple media (such as centralized or distributed database or associated caching and server)." machine can for term Read medium " it will also be considered as including that can store appointing for instruction (for example, the instruction 816) executed by machine (for example, machine 800) The combination of what medium or multiple media so that instruction is in the one or more processors (for example, processor 810) by machine 800 When execution, machine 800 is made to execute any one or more of approach described herein.Therefore, " machine readable media " refers to Individually storage device or equipment and the storage system or storage network of " being based on cloud " including multiple storage devices or equipment.

I/O components 850 may include being used to receive input, providing output, generation output, send information, exchange information, catch Catch the various components of measurement etc..It is included in specific I/O components 850 in specific machine by the type depending on machine. For example, the portable machine of such as mobile phone will likely include touch input device or other such input mechanisms, and nothing Head server machine will likely not include such touch input device.It should be appreciated that I/O components 850 may include in Fig. 8 not The many other components shown.I/O components 850 are grouped according to function, to simplify following discussion, and are grouped not with any Mode is limited.In various example embodiments, I/O components 850 may include output precision 852 and input module 854.It is defeated It may include visual component (for example, display, such as plasma display panel (PDP), light emitting diode to go out component 852 (LED) display, liquid crystal display (LCD), projecting apparatus or cathode-ray tube (CRT)), acoustic assembly (for example, loud speaker), touch Feel component (for example, vibrating motor, resistance mechanisms), alternative signal generator etc..Input module 854 may include that alphanumeric is defeated Enter component (for example, keyboard, the touch screen for being configured to reception alphanumeric input, light-optical keyboard or the input of other alphanumerics Component), the input module based on point is (for example, mouse, touch tablet, trace ball, control stick, motion sensor or another fixed point apparatus Device), sense of touch component (for example, physical button, provide touch or touch gestures position and/or power touch screen or other Sense of touch component), audio input component (for example, microphone) etc..

In other example embodiments, I/O components 850 may include bioassay component 856, moving parts 858, ring Border component 860 or location component 862 and many other components.For example, bioassay component 856 may include for detecting table Up to (for example, wrist-watch reaches, facial expression, phonetic representation, body gesture or eyes tracking), measure bio signal (for example, blood pressure, Heart rate, body temperature, sweat or E.E.G), identification people is (for example, speech recognition, retina identification, face recognition, fingerprint recognition or be based on The identification of electroencephalogram) etc. component.Moving parts 858 may include that acceleration sensing device assembly (for example, accelerometer), gravity pass Sensor component, rotation sensing device assembly (for example, gyroscope) etc..Environment components 860 may include such as illuminance transducer component (example Such as, photometer), the temperature sensor assembly one or more thermometers of environment temperature (for example, detection), humidity sensor group Part, pressure sensor assembly (such as barometer), acoustics sensor device assembly are (for example, one or more wheats of detection ambient noise Gram wind), proximity sensor component (for example, infrared sensor of the neighbouring object of detection), gas sensor (for example, for safety and Detection harmful gas concentration or the gas detection sensor for measuring the pollutant in air) or can provide corresponding to ambient Manage the other assemblies of the instruction of environment, measurement or signal.Location component 862 may include position sensor assembly (for example, the whole world Positioning system (GPS) receiver module), highly sensing device assembly (for example, altimeter or detect air pressure barometer (according to gas Pressure can export height)), orientation sensors component (for example, magnetometer) etc..

Communication can be realized using various technologies.I/O components 850 may include communication component 864, communication set Part 864 is operable machine 800 is coupled to network 880 or equipment 870 via coupling 882 and coupling 872 respectively.For example, logical Letter component 864 may include network interface components or other suitable equipments for being connect with 880 interface of network.In other examples In, communication component 864 may include wire communication component, wireless communication components, cellular communication component, near-field communication (NFC) group Part,Component (such asLow energy consumption),Component and other communicated via the offer of other mode Communication component.Equipment 870 can be any one of another machine or various peripheral equipments (for example, via universal serial bus (USB) peripheral equipment coupled).

In addition, communication component 864 can detect identifier or including operable to detect the component of identifier.For example, logical Believe that component 864 may include radio frequency identification (RFID) tag reader component, NFC intelligent labels detection components, optical pickup Component is (for example, the optical sensor for detecting the following terms：One-dimensional bar code (such as Universial Product Code (UPC) bar shaped Code), multi-dimensional bar code (such as quick response (QR) code), Aztec's code, data matrix, Dataglyph, MaxiCode, PDF417, supersign, UCC RSS-2D bar codes and other optical codes) or Acoustic detection component (for example, mark for identification The microphone of audio signal).Furthermore, it is possible to various information are exported via communication component 864, such as via Internet protocol (IP) position in geographical location, viaThe position of signal triangulation can indicate specific position via detection The position etc. of NFC beacon signals.

Transmission medium

In various example embodiments, one or more parts of network 880 can be self-organizing networks, Intranet, outer Networking, Virtual Private Network (VPN), LAN (LAN), Wireless LAN (WLAN), wide area network (WAN), wireless WAN (WWAN), metropolitan area A part, the plain old telephone service of net (MAN), internet, a part for internet, Public Switched Telephone Network (PSTN) (POTS) network, cellular phone network, wireless network,Network, another type of network or two or more this The combination of the network of sample.For example, a part for network 880 or network 880 may include wireless or cellular network, and couple 882 can be the honeycomb or wireless of CDMA (CDMA) connection, global system for mobile communications (GSM) connection or another type Coupling.In this example, any one of various types of data transmission technologies, such as single carrier may be implemented in coupling 882 Radio transmission techniques (1xRTT), Evolution-Data Optimized (EVDO) technology, General Packet Radio Service (GPRS) technology, GSM Evolution enhances wireless (4G) net of the third generation partner program (3GPP) of data rate (EDGE) technology including 3G, forth generation It is network, Universal Mobile Telecommunications System (UMTS), high-speed packet access (HSPA), World Interoperability for Microwave Access, WiMax (WiMAX), long-term Evolution (LTE) standard, other standards, other remote protocols or other data transmission skills defined by various standard setting organizations Art.

Instruction 816 can make via network interface device (e.g., including the network interface components in communication component 864) It is sent out by network 880 with transmission medium and using multiple well known transport protocols (for example, hypertext transfer protocol (HTTP)) It send or receives.Similarly, can use transmission medium sent to equipment 870 via coupling 872 (for example, equity couple) or from its Receive instruction 816.Term " transmission medium " should be believed to comprise can to store, encode or carry for being executed by machine 800 Any intangible medium of instruction 816, and include communication for promoting the software number analog communication signal or other Intangible medium.Mounting medium includes any medium for carrying machine or computer-readable instruction.Transmission medium and machine readable Jie Matter is the embodiment of mounting medium.

Following numbering example is embodiment.

1. a kind of method, including：

The search inquiry of first language is received from FTP client FTP；

Generate semantic meaning vector associated with described search inquiry；

Access multiple semantic meaning vectors associated with multiple item records, wherein at least some of item record It is not write with first language；

For vectorial with the associated each corresponding semantic meaning of item record：

Corresponding semantic meaning vector sum is inquired associated semantic meaning vector with described search to compare；With And

Based on semantic meaning vector sum associated with item record semantic meaning associated with described search inquiry to Comparison between amount selects one or more item records；

For selected each corresponding item record：

Determine whether corresponding item record is write with first language；And

According to the determination that corresponding item record is not write with first language, automatically corresponding item record is translated For first language；And

One or more selected item records are sent to FTP client FTP for display.

2. according to the method described in example 1, wherein the item record is write with a variety of different language.

3. according to the method described in example 1 or example 2, further include：

Item record is received, for inclusion in network-based business system；

Generate the semantic meaning vector of the item record received；And

Semantic meaning vector is stored in the database of network-based business system.

4. according to the method described in example 3, wherein storing semantic meaning vector further includes：

The item record that analysis joins with semantic meaning vector correlation is to identify the product class joined with semantic meaning vector correlation Not；And

Tissue database so that each semantic meaning vector is associated with the product category determined.

5. the method according to any one of example 1 to 4, wherein by corresponding semantic meaning vector sum and institute It states the associated semantic meaning vector of search inquiry and compares and further include：

Between calculating inquires the associated corresponding semantic meaning vector of semantic meaning vector sum with described search Degree of approach score.

Further include vectorial to multiple semantic meanings based on calculated degree of approach score 6. according to the method described in example 5 It is ranked up.

7. according to the method described in example 6, wherein be based at least partially on the row joined with each semantic meaning vector correlation Sequence selects one or more item records.

8. according to the method described in any one of example 4, example 6 or example 7, wherein access and multiple item record phases Associated multiple semantic meaning vectors further include：

The inquiry of analysis described search is to identify one or more product categories associated with described search inquiry；And

Access semantic meaning vector associated with the one or more product categories identified.

9. according to the method described in any one of example 1 to example 8, wherein generate associated with described search inquiry Semantic meaning vector further includes：

Identification first language associated with described search inquiry；

Semantic meaning vector associated with the first language identified is selected to generate model；And

Model is generated using selected semantic meaning vector to generate the semantic meaning vector of described search inquiry.

10. a kind of system, including：

One or more processors；

Memory；And

Be stored in one or more of memory program, one or more of programs be used for by one or Multiple processors are executed to follow the steps below：

The search inquiry of first language is received from FTP client FTP；

Generate semantic meaning vector associated with described search inquiry；

For selected each corresponding item record：

Determine whether corresponding item record is write with first language；And

One or more selected item records are sent to FTP client FTP for display.

11. according to the system described in example 10, wherein the item record is write with a variety of different language.

12. according to the system described in example 10 or example 11, further include：

Item record is received, for inclusion in network-based business system；

Generate the semantic meaning vector of the item record received；And

13. according to the system described in example 12, wherein storing semantic meaning vector further includes：

14. the system according to any one of example 10 to 13, wherein by corresponding semantic meaning vector sum with Described search inquires associated semantic meaning vector and compares：

Further include being contained to the multiple semanteme based on calculated degree of approach score 15. according to the system described in example 14 Adopted vector is ranked up.

16. a kind of computer readable storage medium of store instruction, described instruction is handled by the one or more of machine Device so that the machine execution includes the operation of the following terms when executing：

The search inquiry of first language is received from FTP client FTP；

Generate semantic meaning vector associated with described search inquiry；

For selected each corresponding item record：

Determine whether corresponding item record is write with first language；And

One or more selected item records are sent to FTP client FTP for display.

17. according to the computer readable storage medium described in example 16, wherein the item record is with a variety of different languages Speech is write.

18. according to the computer readable storage medium described in example 16 or example 17, further include：

Item record is received, for inclusion in network-based business system；

Generate the semantic meaning vector of the item record received；And

19. according to the computer readable storage medium described in example 18, wherein storing semantic meaning vector further includes：

20. the computer readable storage medium according to any one of example 16 to 19, wherein by corresponding language Adopted meaning vector sum and described search are inquired compared with associated semantic meaning vector：

21. a kind of mounting medium of carrying machine readable instructions, the machine readable instructions are by one of machine or more When a processor executes, the machine is made to execute the method according to any one of example 1 to example 9.

Language

Throughout the specification, component, operation or structure as described in single instance may be implemented in plural example.Although The operation of separation is illustrated and is described as in each operation of one or more methods, but one or more of each operation can To be performed simultaneously, and without executing operation in the indicated order.It is illustrated as the structure and function of separation assembly in example arrangement It may be implemented as composite structure or component.Similarly, the structure and function for being illustrated as single component may be implemented as detaching Component.These and other modifications, modification, addition and improvement are fallen into the range of theme herein.

Although describing the general introduction of present subject matter by reference to specific example embodiment, the disclosure is not being departed from In the case of the wider range of embodiment, these embodiments can be carry out various modifications and be changed.Present subject matter these Embodiment can be referred to either individually or collectively by term " invention " herein, merely for the purpose of convenient, and be not intended to certainly Scope of the present application is limited to any single disclosure or inventive concept (if in fact disclosing more than one) by dynamic ground.

The embodiments shown herein fully is described in detail so that those skilled in the art can realize disclosed religion It leads.Other embodiment can be utilized and export other embodiment according to these embodiments, so as to not depart from the disclosure Structure and replacement in logic and change are made in the case of range.Therefore, limit should not be regarded as by being somebody's turn to do " specific implementation mode " Meaning processed, and the range of various embodiments only by the full scope of the equivalent of appended claims and claim come It limits.

As it is used herein, term "or" being interpreted as including property or exclusive meaning.Furthermore, it is possible to be directed to Multiple examples are provided here depicted as the resource of single instance, operation or structure.In addition, various resources, operation, module, drawing The boundary held up between data storage is any to a certain extent, and is shown in the context of specific illustrative configuration Specific operation.Other distribution of function are contemplated, and these distribution can fall into the range of the various embodiments of the disclosure It is interior.In general, in example arrangement as separation resource present structure and function may be implemented as combination structure or Resource.Similarly, the structure and function presented as single resource may be implemented as the resource of separation.These and other become Type, modification, addition and improvement are fallen into the range of the embodiment of the disclosure indicated by appended claims.Therefore, specification It should be seen as illustrative instead of limiting meaning with attached drawing.

Claims

1. a kind of method, including：

The search inquiry of first language is received from FTP client FTP；

Generate semantic meaning vector associated with described search inquiry；

It is vectorial to access multiple semantic meanings associated with multiple item records, wherein at least some of item record is not It is write with first language；

Corresponding semantic meaning vector sum is inquired associated semantic meaning vector with described search to compare；And

Based on semantic meaning vector sum associated with item record semantic meaning vector associated with described search inquiry Between comparison, select one or more item records；

For selected each corresponding item record：

Determine whether corresponding item record is write with first language；And

According to the determination that corresponding item record is not write with first language, corresponding item record is automatically translated as One language；And

One or more selected item records are sent to FTP client FTP for display.

2. according to the method described in claim 1, wherein, the item record is write with a variety of different language.

3. according to the method described in claim 1, further including：

Item record is received, for inclusion in network-based business system；

Generate the semantic meaning vector of the item record received；And

Semantic meaning vector is stored in the database at network-based business system.

4. according to the method described in claim 3, wherein, storage semantic meaning vector further includes：

The item record that analysis joins with semantic meaning vector correlation is to identify the product category joined with semantic meaning vector correlation；With And

5. according to the method described in claim 1, wherein, corresponding semantic meaning vector sum and described search are inquired phase Associated semantic meaning vector, which compares, further includes：

It calculates close between inquiring the associated corresponding semantic meaning vector of semantic meaning vector sum with described search Spend score.

6. according to the method described in claim 5, further including being based on calculated degree of approach score to the multiple semantic meaning Vector is ranked up.

7. according to the method described in claim 6, wherein, being based at least partially on the row joined with each semantic meaning vector correlation Sequence selects one or more item records.

8. according to the method described in claim 4, wherein, accessing multiple semantic meaning vectors associated with multiple item records Further include：

9. according to the method described in claim 1, wherein, generating semantic meaning vector associated with described search inquiry and also wrapping It includes：

Identification first language associated with described search inquiry；

10. a kind of system, including：

One or more processors；

Memory；And

It is stored in one or more of memory program, one or more of programs are used for by one or more of Processor is executed to carry out：

The search inquiry of first language is received from FTP client FTP；

Generate semantic meaning vector associated with described search inquiry；

For selected each corresponding item record：

Determine whether corresponding item record is write with first language；And

One or more selected item records are sent to FTP client FTP for display.

11. system according to claim 10, wherein the item record is write with a variety of different language.

12. system according to claim 10, further includes：

Item record is received, for inclusion in network-based business system；

Generate the semantic meaning vector of the item record received；And

13. system according to claim 12, wherein storing semantic meaning vector further includes：

14. system according to claim 10, wherein inquire corresponding semantic meaning vector sum and described search Associated semantic meaning vector, which compares, further includes：

15. system according to claim 14 further includes being contained to the multiple semanteme based on calculated degree of approach score Adopted vector is ranked up.

16. a kind of computer readable storage medium of store instruction, described instruction are held by the one or more processors of machine Make that the machine execution includes operation below when row：

The search inquiry of first language is received from FTP client FTP；

Generate semantic meaning vector associated with described search inquiry；

For selected each corresponding item record：

Determine whether corresponding item record is write with first language；And

One or more selected item records are sent to FTP client FTP for display.

17. computer readable storage medium according to claim 16, wherein the item record is with a variety of different languages Speech is write.

18. computer readable storage medium according to claim 16, further includes：

Item record is received, for inclusion in network-based business system；

Generate the semantic meaning vector of the item record received；And

19. computer readable storage medium according to claim 18, wherein storing semantic meaning vector further includes：

20. computer readable storage medium according to claim 16, wherein by corresponding semantic meaning vector sum Further include compared with inquiring associated semantic meaning vector with described search：

21. a kind of mounting medium of carrying machine readable instructions, the machine readable instructions are at the one or more by machine When managing device execution, the machine is made to execute method according to any one of claim 1 to 9.